Project BIFROST (Baseline Inferences for Fennoscandian Rebound Observations, Sea-level, and Tectonics) combines networks of continously operating GPS receivers in Sweden and Finland to measure ongoing crustal deformation due to glacial isostatic adjustment (GIA). We present an analysis of data collected between August 1993 and May 2000. We compare the GPS determinations of three-dimensional crustal motion to predictions calculated using the high-resolution Fennoscandian deglaciation model recently proposed by Lambeck et al. [1998a, 1998b]. We find that the maximum observed uplift rate (∼10 mm yr-1) and the maximum predicted uplift rate agree to better than 1 mm yr-1. The patterns of uplift also agree quite well, although significant systematic differences are evident. The root-mean-square residual rate for a linear error model yields estimates of rate accuracy of 0.4 mm yr-1 for east, 0.3 mm yr-1 for north, and 1.3 mm yr-1 for up; these figures incorporate model errors, however. We have also compared the values for the observed radial deformation rates to those based on sea level rates from Baltic tide gauges. The observational error for the vertical GPS rates required to give a reduced X2 of unity is 0.8 mm yr-1. The time series do exhibit temporal variations at seasonal frequencies, as well as apparent low-frequency noise. An empirical orthogonal function analysis indicates that the temporal variations are highly correlated among the sites. The correlation appears to be regional and falls off only slightly with distance. Some of this correlated noise is associated with snow accumulation on the antennas or, for those antennas with radomes, on the radomes. This problem has caused us to modify the radomes used several times, leading to one of our more significant sources of uncertainty.